Chris Dayton
CEO, Co-Founder
Actions
Chris Dayton is the Co-Founder and CEO of QualityAssured.ai, a company that builds closed-system, on-premise AI platforms designed specifically for pharmaceutical and life science environments. His work focuses on compliant model architectures, domain-specific training, and the practical application of large language models in GMP, quality, and regulatory workflows. Chris has led the development of AI systems used for deviation authorship, impact assessments, and risk-based decision support; all with a strong emphasis on traceability, reproducibility, and regulatory alignment. He regularly collaborates with industry experts and regulatory professionals to translate emerging expectations into operational, inspection-ready AI solutions for global pharma and biotech organizations.
Validating the Probablistic: A Risk-Based Framework for AI System Qualification in GMP Environments
AI systems are entering GMP environments at pace — embedded in QMS platforms, laboratory instruments, and deviation management workflows. But the validation frameworks the industry has relied on for decades were not designed for probabilistic, opaque systems that behave differently every time they run. Traditional IQ/OQ/PQ asks: "Does the system calculate correctly every time?" For an LLM-based system, that question has no answer — the output is never exactly the same twice.
This 45-minute presentation introduces a practical, risk-based framework for qualifying AI systems in regulated pharmaceutical environments, grounded in GAMP 5, FDA Computer Software Assurance (CSA) guidance, EMA Annex 22, and the ISPE GAMP AI Good Practice Guide. The core argument is a fundamental shift in validation philosophy: stop trying to verify the algorithm, and start validating the controls around it.
The session covers: why traditional IQ/OQ/PQ is structurally incompatible with probabilistic systems; how to classify AI systems using a risk/autonomy matrix mapped to GAMP 5 Categories and Annex 22 Static/Dynamic designations; the hidden compliance risk of Shadow AI — vendor-enabled AI features running inside validated platforms without QA oversight; and the 4-Layer AI Validation Framework developed from real deployment experience: Configuration Qualification (CQ), Integration & Data Integrity Qualification (IDQ), User Interaction Qualification (UIQ), and Performance Monitoring Qualification (PMQ).
The framework is anchored by a live case study: a production AI system — a Compliance/Gap/Overlap Analyzer — that evaluates SOP sets against a regulatory RAG database. The case study covers the specific validation challenges of a two-corpus AI architecture, confidence score thresholds as acceptance criteria proxies, defect injection methodology for false negative characterization, and how continuous monitoring thresholds and revalidation triggers were established.
Attendees leave with a three-question decision framework applicable to any AI system in their environment: What is the Context of Use and what is the risk if the system is wrong? Does the AI write to a GMP record, influence a GMP decision, or neither? And what controls exist around the output that can be independently validated?
How Do You Validate an AI System? A Hands-On Workshop for Validation Professionals
This 90 minute workshop is designed for validation engineers, quality professionals, and CSV specialists who understand GMP systems validation and want a practical, defensible path forward for AI. No prior AI knowledge is required — the session opens with a plain-language primer on how large language models actually work, what makes them fundamentally different from traditional software, and why that difference matters for validation. Participants are encouraged to ask questions and share their own scenarios throughout.
The session covers: how AI systems work and where they are appearing in regulated environments; why traditional IQ/OQ/PQ is structurally incompatible with probabilistic systems; how to classify an AI system using the Autonomy vs. Design Control matrix mapped to GAMP 5 Categories and EMA Annex 22 designations; how to write a formal Context of Use statement; how to determine 21 CFR Part 11 applicability; how to define acceptance criteria, test cases, and a performance monitoring plan; and what revalidation triggers must be built into your change control process.
The hands-on component asks participants to classify a set of realistic AI system scenarios using the Autonomy vs. Design Control matrix — surfacing the most common scoping errors and grounding the framework in situations the room will recognize from their own organizations.
Participants leave with a takeaway reference guide covering the complete framework: the classification matrix, the Context of Use template, the 4-Layer AI Validation Framework, the Two-Slit Test for 21 CFR Part 11 applicability, and a monitoring plan checklist — tools they can apply immediately back at their organizations.
Compliant-by-Design AI: Architecture Choices and Data Requirements for Regulated Use
Abstract:
This joint session, presented by Chris Dayton (QualityAssured.AI) and Felipe Fontanet (GMP Bridge), integrates architectural best practices with real-world QA and data governance foundations for compliant AI deployment in the life-science industry.
The first portion, led by Chris Dayton, focuses on the architectural decisions that determine whether an AI system can be trusted in regulated environments. Topics include on-prem versus cloud deployment, static versus continuously updated models, lifecycle control, version locking, reproducibility mechanisms, traceability, and the architectural patterns most aligned with emerging expectations in Europe. This section provides a clear and practical framework accessible to both technical and non-technical participants.
The second portion, led by Felipe Fontanet, examines the data foundations required for compliant AI, emphasizing how gold-standard data must be curated, governed, and approved by QA to ensure defensibility. Using examples from EU GMP operations, including deviation narratives, hybrid batch records, environmental monitoring data, and CAPA evidence, this section highlights how data inconsistencies, lineage gaps, and siloed systems often pose greater regulatory risk than the AI model itself.
Together, the session delivers a practical blueprint for designing AI systems where architecture and data handling work together to meet European regulatory expectations.
What attendees will gain from this session:
Attendees will understand the architectural choices that shape AI behavior and regulatory defensibility, including how versioning, reproducibility, and deployment models influence traceability and audit readiness.
They will also gain a practical understanding of data readiness requirements, including how QA defines gold-standard data, how structured and unstructured data must be curated, and how real GMP evidence such as deviations, environmental monitoring, CAPA, and batch records determines whether AI outputs can be defended during inspections.
How attendees will stay engaged:
Clear diagrams, architectural comparisons, and real GMP examples make the material intuitive and immediately applicable. Side-by-side frameworks help attendees understand trade-offs in architecture and data strategy. Short scenarios encourage participants to map concepts to their own environments. The session closes with a Q and A segment to discuss typical challenges and practical considerations for compliant AI adoption.
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top